TUTORIAL 7 – 3D VIDEO CONTRUCTION AND CODING
Objectives:
• To learn how a 3D video is constructed from left- and right-view video components
• To encode the components independently using H26
I. Frame extraction
To extract the left-eye and right-eye views from 3D video, we need to process the video
and separate the red channel (left-eye view) and the cyan channel (right-eye view).
Here is a step-by-step approach for the task:
1. Download the video (from YouTube or any URL).
2. Extract the frames from the video.
3. Separate the channels:
a. Left-eye view: This is typically the red channel.
b. Right-eye view: This is typically the cyan channel (combined green and
blue).
4. Save the left-eye and right-eye images from each frame.
🎬 Full Colab Program to Extract Left-Eye and Right-Eye Views from a 3D
Video
🔧 STEP 1: Install Necessary Dependencies
!pip install yt-dlp opencv-python matplotlib
!pip install imageio
📥 STEP 2: Download the YouTube Video
We'll use yt-dlp to download the YouTube video directly.
import yt_dlp
# Replace the URL with the correct one for the "Roller Coaster Dinosaur T-Rex 3D" video
url = "https://www.youtube.com/watch?v=hdUCJNJgBD8" # Link to the 3D video
output_file = "roller_coaster_3d.mp4"
ydl_opts = {
'format': 'best[ext=mp4]',
'outtmpl': output_file,
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([url])
print(f"Downloaded video as {output_file}")
🎥 STEP 3: Extract Frames from the Video
import cv2
import os
# Create output directories for the left-eye and right-eye views
os.makedirs("left_eye", exist_ok=True)
os.makedirs("right_eye", exist_ok=True)
# Open the video file
cap = cv2.VideoCapture(output_file)
fps = cap.get(cv2.CAP_PROP_FPS)
frame_count = 0
while True:
ret, frame = cap.read()
if not ret:
break
# Split the frame into channels
left_eye_frame = frame.copy() # Left-eye view (red channel)
right_eye_frame = frame.copy() # Right-eye view (cyan)
# Left-eye: Keep only the red channel (typically for the left-eye view)
left_eye_frame[:, :, 1] = 0 # Remove green channel
left_eye_frame[:, :, 2] = 0 # Remove blue channel
# Right-eye: Keep only the green and blue channels (typically for the right-eye view)
right_eye_frame[:, :, 0] = 0 # Remove red channel
# Save the left-eye and right-eye frames as images
cv2.imwrite(f"left_eye/frame_{frame_count:04d}.png", left_eye_frame)
cv2.imwrite(f"right_eye/frame_{frame_count:04d}.png", right_eye_frame)
frame_count += 1
cap.release()
print(f"Extracted {frame_count} frames.")
💾 STEP 4: Download the Left-Eye and Right-Eye Frames
You can download the left-eye and right-eye frames manually or create a zip file to
download the entire set of frames.
import shutil
# Create a zip file for the extracted frames
shutil.make_archive("left_eye_frames", 'zip', "left_eye")
shutil.make_archive("right_eye_frames", 'zip', "right_eye")
# Provide download links
from google.colab import files
files.download("left_eye_frames.zip")
files.download("right_eye_frames.zip")
✅ Summary:
• This program downloads the 3D anaglyph video.
• It extracts frames, then separates the frames into left-eye (red channel) and right-
eye (cyan channel).
• Saves the left-eye and right-eye frames as separate images.
• It generates zip files of the extracted frames, which you can download from Colab.
II. Component compression
To compress the left-eye and right-eye videos into H.264 format using FFmpeg in Google
Colab, we can follow these steps:
Steps:
1. Create the video from extracted frames (for both left-eye and right-eye).
2. Compress both videos using the H.264 codec.
🎬 Full Colab Program to Compress Left-Eye and Right-Eye Views using
H.264
🔧 STEP 1: Install Necessary Dependencies
Make sure you have all dependencies for FFmpeg and video processing installed:
!pip install yt-dlp opencv-python matplotlib
!pip install imageio
📥 STEP 2: Download the YouTube Video (if you haven't already)
import yt_dlp
# Replace with the URL for the "Roller Coaster Dinosaur T-Rex 3D" video
url = "https://www.youtube.com/watch?v=hdUCJNJgBD8" # Link to the 3D video
output_file = "roller_coaster_3d.mp4"
ydl_opts = {
'format': 'best[ext=mp4]',
'outtmpl': output_file,
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([url])
print(f"Downloaded video as {output_file}")
🎥 STEP 3: Extract Frames and Create Left-Eye and Right-Eye Videos
If you've already extracted frames in the previous steps, we can move directly to this
compression part.
Here’s how to create the left-eye and right-eye videos:
import cv2
import os
# Directories for the frames
left_eye_dir = "left_eye"
right_eye_dir = "right_eye"
# Ensure directories exist for video generation
os.makedirs(left_eye_dir, exist_ok=True)
os.makedirs(right_eye_dir, exist_ok=True)
# Open the original 3D video for frame extraction
cap = cv2.VideoCapture(output_file)
fps = cap.get(cv2.CAP_PROP_FPS)
frame_count = 0
# Create left-eye and right-eye view frames
while True:
ret, frame = cap.read()
if not ret:
break
left_eye_frame = frame.copy()
right_eye_frame = frame.copy()
# Left-eye: Keep only the red channel
left_eye_frame[:, :, 1] = 0
left_eye_frame[:, :, 2] = 0
# Right-eye: Keep only the green and blue channels
right_eye_frame[:, :, 0] = 0
# Save the left-eye and right-eye frames
cv2.imwrite(f"{left_eye_dir}/frame_{frame_count:04d}.png", left_eye_frame)
cv2.imwrite(f"{right_eye_dir}/frame_{frame_count:04d}.png", right_eye_frame)
frame_count += 1
cap.release()
print(f"Extracted {frame_count} frames.")
🖥️ STEP 4: Create Left-Eye and Right-Eye Videos
Now, we use FFmpeg to compress the frames and generate H.264 encoded videos.
# Create left-eye and right-eye videos using H.264 compression
!ffmpeg -framerate {fps} -i {left_eye_dir}/frame_%04d.png -c:v libx264 -pix_fmt yuv420p
left_eye_video.mp4 -y
!ffmpeg -framerate {fps} -i {right_eye_dir}/frame_%04d.png -c:v libx264 -pix_fmt yuv420p
right_eye_video.mp4 -y
📤 STEP 5: Download the Compressed Videos
Finally, download the compressed left-eye and right-eye videos:
from google.colab import files
files.download("left_eye_video.mp4")
files.download("right_eye_video.mp4")
✅ Summary
• This script downloads the YouTube video, extracts the frames, separates them into
left-eye and right-eye views, and then compresses them into H.264 video format.
• You can download both the left-eye and right-eye videos for viewing with red/cyan
3D glasses.
• The compression process uses FFmpeg with the libx264 codec.
